Pricor
A locally run application for parsing supplier quotation sheets and computing costs. It combines a fine-tuned language model with a deterministic calculation engine to convert Excel/CSV spreadsheets into structured product records, emphasizing auditability and user confirmation.
Timeline (last 90 days)
Pricor is a local app that converts Excel or CSV files into consistently structured, importable product records containing supplier code, description, net unit cost and currency, while preserving row-level traceability.
Pricor's core design principle is: the model understands, the engine computes, the user decides.
Pricor's executor only understands a closed set of operations (parsing numbers, applying chained discounts, removing VAT, splitting packaging, mapping values, normalizing text, optional currency conversion), never executes model-generated code, and never hands final price calculation to the model.
Pricor uses the Decimal (decimal fixed-point) type for parsing, discounts, tax removal and splitting to avoid the accumulation of IEEE 754 floating-point errors.
Pricor's processing pipeline is: first a rule analyzer finds the header row, row types, column types and number formats, then the fine-tuned multilingual model Laya assigns each column to a closed set of roles, columns Laya is unsure about go to Gemma 4 E4B running locally via Ollama, and anything still uncertain is handed to the user to answer.
Pricor's memory mechanism creates memories only when the user actively writes one or accepts a Gemma suggestion and clicks "remember this," never automatically; the relevant memories are passed to Gemma as numbered untrusted data, Gemma must cite the memories it relies on, and citing a memory that was not provided causes the entire reply to be discarded.
Pricor's published benchmark uses 29 held-out columns from two distributors: keyword/type heuristics were correct on 24/29 (82.8%), Laya zero-shot on 18/29 (62.1%), fine-tuned Laya on 22/29 (75.9%), and pure prompting of Gemma on 25/29 (86.2%).
In the full-table cascade, only 11 of the 29 columns were resolved automatically (10 of them correctly), and 18 columns still required user intervention.
Pricor has a percentage semantics bug: Excel stores a cell displayed as 21% as 0.21, and once the number format is lost, a 21% VAT becomes 0.21%, yet it still computes a cost that looks reasonable.
Pricor has a second discount column bug: when a table has two independent discount columns, the plan builder only uses the first, which violates the app's own "never silent" rule, and a fix is on the to-do list.
1 more timeline events
All Facts (11)
Pricor is a local app that converts Excel or CSV files into consistently structured, importable product records containing supplier code, description, net unit cost and currency, while preserving row-level traceability.
50%UnverifiedPricor's core design principle is: the model understands, the engine computes, the user decides.
50%UnverifiedPricor's executor only understands a closed set of operations (parsing numbers, applying chained discounts, removing VAT, splitting packaging, mapping values, normalizing text, optional currency conversion), never executes model-generated code, and never hands final price calculation to the model.
50%UnverifiedPricor uses the Decimal (decimal fixed-point) type for parsing, discounts, tax removal and splitting to avoid the accumulation of IEEE 754 floating-point errors.
50%UnverifiedPricor's processing pipeline is: first a rule analyzer finds the header row, row types, column types and number formats, then the fine-tuned multilingual model Laya assigns each column to a closed set of roles, columns Laya is unsure about go to Gemma 4 E4B running locally via Ollama, and anything still uncertain is handed to the user to answer.
50%UnverifiedPricor's memory mechanism creates memories only when the user actively writes one or accepts a Gemma suggestion and clicks "remember this," never automatically; the relevant memories are passed to Gemma as numbered untrusted data, Gemma must cite the memories it relies on, and citing a memory that was not provided causes the entire reply to be discarded.
50%UnverifiedPricor's published benchmark uses 29 held-out columns from two distributors: keyword/type heuristics were correct on 24/29 (82.8%), Laya zero-shot on 18/29 (62.1%), fine-tuned Laya on 22/29 (75.9%), and pure prompting of Gemma on 25/29 (86.2%).
50%UnverifiedIn the full-table cascade, only 11 of the 29 columns were resolved automatically (10 of them correctly), and 18 columns still required user intervention.
50%UnverifiedPricor has a percentage semantics bug: Excel stores a cell displayed as 21% as 0.21, and once the number format is lost, a 21% VAT becomes 0.21%, yet it still computes a cost that looks reasonable.
50%UnverifiedPricor has a second discount column bug: when a table has two independent discount columns, the plan builder only uses the first, which violates the app's own "never silent" rule, and a fix is on the to-do list.
50%UnverifiedPricor's backend test suite passes 1,219 tests and the frontend passes 194 tests.
50%